Pilot lab: UI generation from a brief without falling into generic patterns
A seeded design-and-implementation lab for judging whether a coding agent can translate a product brief into intentional interface choices.
The product layer matters when the workflow can preserve design intent across multiple edits instead of regressing into safe defaults.
Model quality shows up in layout judgment, hierarchy, and whether the system can translate abstract direction into concrete UI decisions.
The output counts only if the result feels deliberate and can survive iteration without the human redesigning it by hand.
Frontend product task with explicit design direction, responsive constraints, and a requirement to avoid generic dashboard aesthetics.
Build a focused landing page from a written brief, choose a coherent visual language, and make the result responsive without collapsing into boilerplate UI.
Why this belongs in the lab suite
Many coding-agent comparisons overfocus on backend or benchmark tasks and ignore UI generation, even though design translation is a real developer workflow.
This lab should ask whether the system can:
- build hierarchy from prose,
- choose a clear visual direction,
- and preserve that direction while implementing responsive behavior.
Failure condition
If the page technically works but feels interchangeable with any other AI landing page, the report should count that as a real quality miss.